Alex Sacerdote on S-Curves, the AI Boom, and Finding Technology Winners

Alex Sacerdote, founder of Whale Rock Capital Management, has spent two decades investing through technology platform shifts — mobile, cloud, e-commerce — by tracking where a company sits on its adoption S-curve. He argues enterprise AI adoption is not following the usual S-curve shape at all: it is going ‘straight up’, which is why Whale Rock calls it an L-curve.

Key ideas

1. Anthropic became Whale Rock’s highest-conviction position by outlasting a field of fifty. Whale Rock began with the infrastructure layer after ChatGPT’s November 2022 launch, reasoning that whoever won the foundational-model race would need enormous compute regardless. Over the following three years, dozens of well-funded challengers — including efforts inside Amazon and Meta — fell away or faltered, leaving what Sacerdote sees as a three-horse oligopoly of Anthropic, OpenAI, and Google, structurally similar to how three cloud providers came to underpin the SaaS market.
2. Code was the true unlock, not a side feature. Whale Rock tracked coding-tool spend as a live signal of AI’s revenue potential: from a $20-a-month grammar-checker era (GitHub Copilot) to reports of individual engineers spending $100 a day on tokens inside Anthropic — implying $20-30k a year per coder, and a market worth roughly half a trillion dollars across the world’s 20 million coders on technology that was, at the time, only seven or eight months old.
3. Enterprise AI adoption is an ‘L-curve’, not an S-curve. Sacerdote puts enterprise application AI at under 1% penetration and knowledge-worker AI usage at roughly 10 basis points — the earliest tinkerer stage of a classic adoption curve — yet expects it to reach 15% within four years. Because AI needs no integration (‘you just open up the browser and it’s there’), it skips the slow, plugged-into-existing-systems adoption pattern of dishwashers and B2B software and goes straight up instead.
4. The investment framework has three legs: S-curve, competitive advantage, underappreciated earnings power. Whale Rock looks for a strong technology adoption curve, a durable moat, and earnings that the market is not yet pricing as exponential — the combination that let the firm buy Nvidia at four times earnings in 2023, Tesla at five times in 2019, and Apple at four times in its early years, because the market persistently fails to think exponentially about compounding businesses.
5. AI foundational models turned out to be far less commoditised than expected. Anthropic, OpenAI, and Google have developed distinct technical specialisms (Anthropic for finance and private equity work, Google for ingesting PDFs), and Anthropic is building a full ecosystem — SDK, orchestration layer, and a ‘harness’ of tools around the raw API — the same lock-in pattern Whale Rock first saw AWS build in 2013 around what looked at the time like commodity servers.
6. Whale Rock sold almost all its enterprise software holdings, moving net short entering this year. Incumbent software companies’ early AI features were not moving the needle and could not be monetised, while CIOs redirected budget toward frontier-model tokens instead. Sacerdote’s revised ‘rule of 40’ for the AI era multiplies a company’s percentage of AI-driven revenue by its market share in that category — and most software incumbents score in the low single digits on the first term.

Content

Spotting the inflection before the data shows it

Sacerdote treats the early, flat part of an S-curve as a pattern-recognition problem rather than a data problem — echoing Andy Grove’s observation that strategic inflection points have to be read through intuition and anecdote before the numbers confirm them. He cites watching a Chinese schoolboy play an advanced game on a large-screen phone as the visual cue that mobile gaming was about to take off, and watching a Gartner IT Symposium ballroom fill to capacity hour after hour as the tell for cloud’s inflection. Adoption speed itself varies by category: consumer technologies like radio can reach saturation in years, while anything that has to be ‘plugged into the back end’ of existing systems — a dishwasher, B2B software — takes far longer, which is the risk he flags for enterprise AI if security and cultural resistance slow it down the way they once slowed cloud adoption.

What could go wrong

Sacerdote names three risks to the bull case: public and political sentiment (he cites Maine banning data centres, and survey data showing only 20% of the public optimistic about AI); a plateau in model quality that would let open-source catch up and turn pricing into a race to the bottom, which would hurt model-layer economics even as it helped chip demand; and the possibility that a major lab falters and its compute commitments go unfilled — though he notes that when Oracle cancelled a large deal, Meta absorbed the capacity almost immediately.

See also

• Anthony Scilipoti on Forensic Accounting, the AI Bubble, and Spotting Corporate Collapse — the bear case on the same boom, argued from financial statements and circular financing rather than adoption curves